[00:00] Hello everyone, and welcome back to N3uron Academy. In the previous video, we focused on MCP Server Tools. Today, we will continue with the other two main MCP primitives: Prompts and Resources,again using AI-Ready PV Demo project. Together with Tools, they provide actions, reusable instructions, and controlled context from the N3uron node. Not every MCP client currently exposes the complete MCP specification; many are still mainly focused on Tools. Support for prompts, resources, resource templates, and subscriptions can therefore vary between clients. N3uron implements the three main MCP primitives, and we continuously follow the protocol as it evolves. As with Tools, the access token determines which Prompts, Resources, and tool categories a client is allowed to use. Let’s start with Prompts. A custom prompt is a reusable instruction template stored directly in the N3uron node.

[01:02] Instead of rewriting the same operational request every time, we define the workflow once and expose it to permitted clients. Each prompt has a name, title, description, and prompt body. The body contains the reusable instructions delivered to the model and can describe the workflow, expected reasoning, or tools to use. Prompts can also define arguments, which are referenced in the prompt body using double angle brackets. This makes the prompt reusable at node level. The workflow remains the same, while only values such as the plant path, time window, or threshold need to change for each invocation. Now let’s look at Resources. Resources expose readable context through MCP, and N3uron also lets us publish custom embedded files from the MCP Server configuration. In this demo, we include the Xantrex GT500 inverter documentation as a custom resource. The same approach can be used for datasheets, operating procedures, company rules, engineering specifications, maintenance guides, or other reference information.

[02:01] A resource is not itself a system prompt; it is controlled reference context that an authorized client can retrieve when needed. Now we switch to MCP Inspector. Here I use the local Streamable HTTP endpoint and provide the authorization information required by the N3uron access token. Once the connection is established, Inspector can discover only the capabilities permitted for that token. The connection is now active, and we can inspect the MCP primitives exposed by the server. We first open Prompts and list the prompts published by the node. I select the audit configuration prompt and provide its arguments, including the plant path and the maximum acceptable backup age. When I request the prompt, N3uron resolves the argument placeholders and returns the final instruction text to the client. Next, I select the find worst assets prompt. This prompt defines three arguments: the lookback period, the plant path, and the underperformance threshold. After entering the values, Inspector shows the resolved prompt, including the tool call and the arguments that the workflow expects.

[03:04] Changing an argument produces a new resolved prompt without changing the reusable template stored in N3uron. That separation is what makes prompt arguments useful for repeatable workflows across different requests. Back in Resources, we list the resources exposed by MCP Server. Among them are the N3uron log files, which can be opened and read directly by a compatible MCP client. Here I briefly return to the N3uron Diagnostics view. Because these operational logs can also be exposed through MCP, they become useful context for troubleshooting and root-cause analysis. Back in Inspector, we continue with Resources and open the custom Xantrex GT500 document. The document is maintained once in the node and can be retrieved by authorized clients that support MCP Resources. This is useful for manuals, datasheets, procedures, specifications, or company rules that should travel with the operational context. N3uron also exposes Resource Templates. In this case, the available Resource Templates are tag-template and alarm-template.

[04:03] The tag-template lets the client provide a tag path and resolve that specific N3uron tag as a resource. After selecting a tag path, Inspector reads the generated resource and returns the current tag information. Because the resource is subscribable, a client that supports subscriptions can also receive update notifications when the resource changes. The alarm-template follows the same idea for alarm information. We select an alarm path, read the generated alarm resource, and obtain its current state and metadata through MCP. We can also subscribe to that alarm resource so the server can notify the client when the underlying state changes, instead of relying only on polling. Here we briefly return to N3uron to inspect the underlying operational information that is being exposed through the resource interface. Back in Inspector, the Server Notifications panel shows the resource update notifications received from N3uron. So Resources can represent static documentation as well as live operational context, while Resource Templates make tag and alarm access reusable across many paths.

[05:07] As always, subscription behavior also depends on whether the MCP client implements that part of the specification. Now let’s move to Claude Desktop and use the same N3uron connection from an AI client. From the connector menu, I can see the Prompts and Resources exposed by N3uron. I first select the Xantrex GT500 resource, preview it, and attach it as reference context for the task. Next, I select the find worst assets prompt from the same connector. Claude asks only for the arguments defined by that prompt. I provide a thirty-day lookback, the BLUELAKE plant path, and a three-percent threshold. The analysis procedure itself does not need to be rewritten, because it is already stored in N3uron. Claude now shows the prompt resolved by N3uron with those values. At this point, both the reusable prompt and the Xantrex reference document are attached to the task. I simply ask Claude to execute the provided prompt.

[06:02] The prompt gives the model a structured starting point, and Claude can then call the permitted N3uron tools required to complete the workflow. Once the first analysis completes, Claude identifies the weakest assets and summarizes the findings. I then ask it to continue with additional operational evidence, including alarm history and downtime against irradiance. Next, I request a seven-day maintenance plan using the Xantrex manual as supporting context.

[07:33] Once complete, Claude combines the operational evidence gathered through N3uron with the technical information from the Resource. The maintenance plan prioritizes the affected inverters and proposes checks and actions grounded in both plant data and inverter documentation. This is a practical example of the three MCP primitives working together. Prompts provide the reusable workflow, Tools perform controlled operations, and Resources provide supporting context.

[08:00] The model can combine those elements with other available context and reasoning, while access remains governed by MCP Server. Finally, we return to N3uron and inspect the plant historical data information directly. We will check whether the inverters reported as underperforming are indeed underperforming. The analysis confirms that this is the case at Power Stations 3 and 6. The AI result does not replace the underlying industrial data; N3uron remains the controlled source for real-time values, history, alarms, logs, and the MCP capabilities exposed to the client. Looking at the historical trends gives us another way to verify the conclusions from the analysis. Keeping plant data, deterministic Tools, reusable Prompts, and reference Resources as separate layers makes the workflow easier to understand, reuse, and verify. As always, you can find further information in the N3uron Knowledge Base. The MCP Server documentation covers Prompts and their arguments, custom and built-in Resources, access control, Resource Templates, and subscriptions. Remember that the exact experience depends on the MCP client, because not every client currently implements every part of the specification.

[09:07] N3uron continues to follow the evolution of MCP and incorporate protocol updates as the ecosystem develops. Thanks for watching. Take care, and see you in the next N3uron Academy video..